A Velocity-Representation Model For MT Cells Eero Simoncelli - - PowerPoint PPT Presentation

▶

Oct 24, 2023 382 likes •528 views

A Velocity-Representation Model For MT Cells Eero Simoncelli Computer and Information Science Department University of Pennsylvania David Heeger Psychology Department Stanford University ARVO, 94 0 Input: normalized STE Input: image

SLIDE 1

A Velocity-Representation Model For MT Cells

Eero Simoncelli Computer and Information Science Department University of Pennsylvania David Heeger Psychology Department Stanford University

ARVO, ’94

SLIDE 2

+

. . .

. . . . . .

σe-2 Input: normalized STE Input: image intensities Output: normalized STE σv-2 Output: normalized VE

. . .

. . .

. . . . . . +

αv

. . .

Two stages of computation, corresponding to:

1. V1 “simple” cells (tuned for Spatio-Temporal Energy).
2. MT “pattern” cells (tuned for “Velocity Energy”).

ARVO, ’94 1

SLIDE 3

Stage I

Space-Time Weighting Fourier Tuning

x y t

ωx ωy ωt

Combines image intensities via space-time oriented linear Þlters. Tuning: localized in the Fourier domain. A population of these mechanisms provides a distributed represen-

tation of spatio-temporal energy (STE).

ARVO, ’94 2

SLIDE 4

Stage I is Not Velocity-Tuned

ωx ωy ωt

Power spectrum of a translating 2D pattern lies on a tilted

plane.

A model V1 unit will respond to many such planes.

ARVO, ’94 3

SLIDE 5

Stage II

STE weighting Velocity Tuning ωx ωy ωt

vx vy

Combines outputs of V1 units tuned for different orientations. Tuning: localized in the image-velocity domain. A population of these mechanisms provides a distributed represen-

tation of velocity.

ARVO, ’94 4

SLIDE 6

Model Parameters

1. Stage I:

“Coverage” of Fourier domain. “Order” of Þlters. Semi-saturation constant,

e.
2. Stage II:

“Coverage” of velocity domain. Resting Þring rate,

Semi-saturation constant,

ARVO, ’94 5

SLIDE 7

Direction Tuning: Stage I

Movshon et al., 1983 Model gratings plaids

ARVO, ’94 6

SLIDE 8

Direction Tuning: Stage II

Movshon et al., 1983 Model gratings plaids

ARVO, ’94 7

SLIDE 9

Speed Tuning

1/16 1/4 1 4 50 100 .5 1 1/16 1/4 1 4

Response Speed Relative to Optimum Maunsell & Van Essen (1983) Model

spontaneous rate

ARVO, ’94 8

SLIDE 10

Correlation Response Function

25 50 75 100 50 100 150

Britten et al. (1993) Response

.25 .5 .75 25 50 75 100

Model Stimulus Correlation (%)

spontaneous rate non-preferred direction preferred direction

ARVO, ’94 9

SLIDE 11

Model Snowden et al. (1991) Response

64 128 192 256 40 80 64 128 192 256 .5 1

Number of dots in preferred direction

Number of dots in non-preferred direction spontaneous rate 16 64 256

ARVO, ’94 10

Response vs. Number of Dots

SLIDE 12

Non-Preferred Suppression

90 180 .5 1

Model

90 180

Snowden et al. (1991) Response Mask Direction

30 60 30 60 30 60 spontaneous rate preferred + mask preferred alone

ARVO, ’94 11

SLIDE 13

Model Behavior: Gratings

Stage I Stage II Mean response plotted as a function of grating normal velocity.

ARVO, ’94 12

SLIDE 14

Model Behavior: Random Dots

Stage I Stage II Mean response plotted as a function of dot drift velocity.

ARVO, ’94 13